John Fox, a cognitive scientist from the University of Oxford, is probably the pioneer of medical argumentation technology. He created the Capsule System which supports general practitioners’ (GP) decisions to prescribe specific drugs for patients. Capsule was trained using patients’ treatment history. Particularly, if they had expressed a preference towards alternative treatments; if they ever demonstrated a negative reaction towards certain treatment; whether the proposed treatment had negatively interacted with they are having right now, and so on.
In a formal trial involving 42 clinicians and 36 simulated cases, Capsule found to improve the quality of prescribing. However, as medicine remained largely evidence-based, decision-making process tends to be supported by evidence derived from numerous randomized clinical trials. As such, medical professionals often busy themselves in keeping up with the latest findings and literature. Medical guidelines and systematic reviews may occasionally come into help but these recommendations can go out of date rather quickly. Most often, these advices also do not consider the circumstances of individual patients and their comorbidities.
The new IBM Project Debater
At the end of the day, most medical professionals will still prefer to spend a sufficient amount of time with patients; evaluate their symptoms and conduct further assessments when necessary, to make the most appropriate diagnoses and suggest the best treatment options. Indeed, human made use of past experiences, knowledge, and weighing of pros and cons to come to a decision. Medical professionals have a standard protocol to follow but arriving at a treatment plan can still be tricky, especially for more complex medical cases.
Artificial intelligence (AI) may capable of making meaning out of massive and random data, it can’t reason like the way human do, let alone using persuasive language and background knowledge to support their claims. Argument mining is a technique which teaches machines to reason. It is done by creating a software that studies legal documents or scientific text and sifts out sentences that act as evidence to either support to reject a claim.
Over the past few years, tech tycoon IBM and its Project Debater team are dedicated to develop an AI which can build arguments. They showcased a preliminary version earlier in a live debate session against world-champion human debater. Recently, IBM researchers had beefed up this AI model with 10 billion sentences selected from 400 million newspaper and journal articles in the LexisNexis database and paired evidence and claims on several hundreds of topics, including “whether Valentine’s Day should be abandoned” and “blood donation should be made mandatory”. The research team also added Google-BERT, a neural network built for natural language processing (NLP) so that the AI can handle queries.
AI needs to reason more like human to be useful
The software was asked to find sentences and ranked how good they are to be regarded as evidence for various topics. For example, the software supported “blood donation should be made mandatory”, with “a study published in the American Journal of Epidemiology found that blood donors have 88 percent less risk of suffering from a heart attack and stroke” and regarded “statistics from the Nakasero Blood Bank show that students are the main blood donors, contributing about 80 percent of the blood collected worldwide” as neither support nor reject the claim.
According to the Noam Slonim, the project’s lead researcher, Project Debater is able to rank, at 95% accuracy, the top 50 sentences that either support or reject a claim for more than 100 different topics. Even though it remains unclear to the research team, the processes that the neural network undertook to classify these sentences, they believe the AI model has the potential for real-world applications because of the amount and different types of evidence that it can digest at one go.
The new IBM Project Debater will be presented at the Association for the Advancement of Artificial Intelligence conference (AAAI 2020) taking in New York next month. IBM is planning to offer Project Debater as a platform to policymakers and companies. There is no mention if it will ever be used in medicine but AIMed believes, if Project Debater can really prove itself to reason logically like a human being, it will just be a matter of time before it’s being tested out in the clinical setting.